【摘要】 目的 提出一种基于主成分分析（principal component analysis, PCA）的深度前馈神经网络（deep feedforward neural network, DFNN）,建立一个适用于中国慢性肾脏病（chronic kidney disease, CKD）人群的肾小球滤过率（glomerular filtration rate, GFR）估算模型，并探讨其在慢性肾脏病患者肾小球滤过率估算中的应用。方法 受试者为2019年5月—2021年1月就诊于安徽医科大学第二附属医院，排除年龄<18岁的肾功能不稳定，服用甲氧苄啶或西咪替丁或接受透析后的163例患者。本研究以^99mTc-DTPA肾动态显像测定GFR为标准，建立主成分分析的深度前馈神经网络（deep feedforward neural network, DFNN）模型，以此估算GFR,同时将估算GFR结果与传统CG方程和BP神经网络估算结果进行对比分析。结果 通过PCA-DFNN-1神经网络训练出来的估算模型的15%符合率、30%符合率、50%符合率分别为38.77%、55.1%、75.5%;ROC曲线下面积为0.845;Youden指数为0.58。结论 提出的基于主成分分析的深度前馈神经网络模型有优于CG方程和BP神经网络模型的结果，可以用于估算GFR。更多还原

【Abstract】 Objective A deep feedforward neural network based on principal component analysis was proposed to establish a glomerular filtration rate estimation model suitable for Chinese chronic kidney disease population, and to explore its use in the estimation of glomerular filtration rate(GFR) in chronic kidney disease patients. Methods The participants were 163 patients who visited The Second Hospital of Anhui Medical University from May 2019 to January 2021,excluding the patients under 18 years old with unstable renal function, taking trimethoprim or crimetidine or receiving dialysis. In this study, the GFR was determined by dynamic renal imaging as the standard, and a deep feedforward neural network(DFNN) model based on principal component analysis was established to estimate GFR. Results The 15%,30%,and 50% coincidence rates of the estimated models trained by the PCA-DFNN-1 neural network were 38.77%,55.1%,and 75.5%; the area under the ROC curve was 0.845; the Youden index was 0.58. Conclusions The proposed deep feedforward neural network model based on principal component analysis has better results than CG equation and BP neural network model, and can be used to estimate GFR.更多还原